Freelancer profile translated to English.

Description

I hold a PhD in AI with over 7 years of experience transforming cutting-edge research into concrete, robust, and high-impact business AI products.

My expertise lies at the intersection of LLMs, VLMs, computer vision, and biometrics, with a particular focus on facial recognition, applied deep learning, and production-ready AI systems. I have worked in both R&D and product/engineering environments, helping teams transition from "nice prototypes" to scalable and reliable AI solutions.

I have contributed to and led advanced projects involving LLMs, RAG systems, model evaluation, ML system design, and vision pipelines. I am recognized for bridging scientific excellence with practical delivery: I don't just build models; I help define the right architecture, trade-offs, and strategy for AI to truly function in production.

Today, I serve as a Senior / Principal Consultant & AI Architect, assisting companies with:

Designing and auditing AI / ML architectures

Building or improving LLM and vision systems

Reducing risks on complex ML projects

Transforming research prototypes into reliable products

Structuring their AI roadmap and upskilling teams

Clients choose me for three reasons: technical credibility, strategic clarity, and execution pragmatism. If you need someone who can both challenge your technical choices and help you deliver faster and more reliably, you've come to the right place.

Typical engagements: AI architecture, LLM systems, vision & biometrics, technical strategy, model audits, R&D acceleration, team mentoring.

Industry field of expertise

Languages

French
Native or bilingual
English
Fluent
Arabic
Native or bilingual

Workplace preferences

Can work on-site

Paris (up to 50km)

Thales
AI Applied research scientist
January 2022 - Today (4 years and 7 months)
Meudon, France
Developed and optimized biometric algorithms, ensuring they met predefined performance targets during rigorous internal and external evaluations

Implemented, developed, and integrated advanced algorithms into various internal and external deliverables, including Software Development Kits, software system solutions, and hardware products.

Established scalable AI pipelines: Automated data preprocessing, model training, and evaluation workflows, leveraging PyTorch and TensorFlow for production-grade model delivery

Supported, tested, and troubleshooted algorithms for existing and new technologies and products, ensuring high accuracy and speed in biometric identification
AWS CUDA C/C++ Pytorch Biometrics
Nokia Bell Labs
PhD student/Research engineer
TELECOMMUNICATIONS
October 2018 - October 2021 (3 years)
Massy, France
Explored the potential of Graph Neural Networks (GNNs) and deep reinforcement learning in addressing the resource allocation problem in 5G network slicing.
- Unveiled the capability of GNNs and deep reinforcement learning in solving resource allocation challenges within 5G network slicing.
- Implemented a reinforcement learning-based strategy to enhance the performance of heuristics, focusing on gap optimality, for the resource allocation problem in 5G network slicing.
- Proposed a machine learning-based solution for monitoring 5G network slices, enabling efficient management of resources.
- Presented a novel learning approach combining graph neural network models and genetic algorithms to determine the optimal monitoring of network slices with probing cycles in 5G.
- Investigated the application of deep reinforcement learning, graph neural networks, and combinatorial optimization in the context of 5G network slicing.
- Contributed to advancing the understanding of optimized resource management techniques and their application in future 5G networks.
Deep Learning Machine learning Reinforcement Learning Pytorch C/C++
I3S
Deep/Machine learning
HEALTH AND WELLNESS
March 2018 - August 2018 (6 months)
Nice, France
- Developed and trained machine and deep learning algorithms on electronic health records (EHR) for improved performance.
- Implemented strategies to parallelize the algorithms on a Jetson TX2 cluster, optimizing for low-power embedded modules within hospitals.
- Balanced model precision, execution time, and memory usage, achieving a trade-off that met the unique requirements of the healthcare environment and reduce the energy consumption of training and inference.
- Utilized TensorFlow, scikit-learn, and Keras to optimize and fine-tune the learning models.
- Leveraged distributed and parallel deep learning techniques to process and analyze large volumes of medical data efficiently.
- Employed NVIDIA Jetson TX2 and CUDA to accelerate the training process and enhance overall performance.
- Successfully achieved a compromise between accuracy, execution time, and memory usage for practical deployment in hospitals.
Deep Learning TensorFlow Keras